12 questions to ask any vendor about AI in property management
Before you let any software's AI touch your residents, owners or bank accounts, ask who turns it on, what it can do alone, what always waits for a person, how you see and stop it, and what happens to your data. Here are 12 questions and what a good answer sounds like.
Every property management vendor now says its product has AI. The phrase covers everything from a button that rewrites an email to agents that answer leads, book vendors and pay bills. A demo will show you the best case. These questions get at the controls, which matter more than the best case. Take them to any vendor, including us. Ask for the answer in the product, not on a slide.
1. Is the AI off until we turn it on, and who can turn it on?
A good answer: it is off by default, and only an admin can turn it on, area by area. Be wary if AI features are on for every user from day one and you have to find the settings to turn them off.
2. Can we set the AI separately for each area of work?
Leasing, collections, maintenance and bookkeeping carry very different risks. A good answer lets you set each area separately, for example drafting only in collections while it handles routine maintenance requests on its own. One switch for all AI is a warning sign.
3. What can it do with money, and what are the limits?
Ask for the exact limits: per step, per day, by property. Ask what happens when a step goes over a limit. A good answer: the step stops and waits for a person, with the reason shown. Ask to see the setting screen.
4. What always waits for a person, no matter how we set it?
There should be a list. Examples of reasonable items: assigning a vendor, sending an account to a collections agency, paying an owner, anything that carries legal weight. If the vendor says everything is configurable, ask what stops a user from configuring something risky.
5. When the AI acts, whose permissions does it use?
A good answer: it acts as a named person, with that person's permissions, and can't do anything that person couldn't. When you approve a step, it should run as you, with your permissions. Be careful with any system where the AI has its own super-user access.
6. Is there a log of everything the AI did?
Ask to see it. A good log shows each step, when it ran, whose authority it used, what it changed and whether it worked. Ideally it is separate from the staff audit log, so you can review AI work on its own. Ask whether the staff audit log can be edited. Append-only is the answer you want.
7. How do we stop it, all at once?
There should be one switch that stops every AI area immediately, plus a switch per area. Ask who can use it and what happens to work in progress. Ask the vendor to show you, live.
8. Can we try it on real work without it acting?
A trial mode where the AI drafts everything but a person approves every message and change is the safest way to judge quality on your own data. Ask how long you can stay in that mode. The right answer is as long as you like.
9. What happens when the AI fails?
Every system fails sometimes. Ask how a failure is shown. A good answer: failed steps are listed plainly, nothing is half-changed, and you can redo the step by hand or retry. Silent failures are the dangerous kind.
10. Is our data used to train AI models?
Ask for the vendor's written policy, and ask about the AI provider underneath it. Ask whether your residents' messages and your financial data leave the platform, and where they go. Also ask how your data is kept separate from other companies' data. If you have questions about privacy obligations to your residents, take them to counsel. This is not legal advice.
11. How does it handle messages to residents and applicants?
Messages to people are where an AI can create the most trouble with the least money involved. Ask which messages it sends on its own, which wait, and whether anything about money, leases, notices or emergencies always waits. Ask how it treats fair housing in leasing conversations and what checks run before a message goes out. Your own fair housing obligations don't change because software wrote the message. This is not legal advice.
12. What does the AI cost to run, and can we cap it?
Some vendors charge per conversation, per unit or per action. Ask how AI usage is billed and whether you can set a hard monthly cap on AI spend. A budget cap is a simple control that protects you from a runaway job.
How to score the answers
Give each question one of three marks: shown in the product, described but not shown, or not available. Anything about money, logs or stopping the AI should be shown in the product before you sign. A vendor that can show you all 12 is rare. A vendor that can show you the money limits, the log and the stop switch is worth a second meeting.
Questions to ask your own team
- Who will own the AI settings, and who reviews the log each week?
- What are our staff approval limits today, by area? The AI's limits should start below them.
- Which properties have clean enough data to start with?
- Who do we tell, inside and outside the company, if we stop the AI?
How Zrico answers these questions
For transparency, here are our own answers. Zrico AI is off until an admin turns it on. Each area is set to Off, Assist or Autopilot. Money caps work per step and per day and can be limited to chosen properties. Legal steps are off unless allowed, and some steps always wait for a person, such as assigning a vendor. Approving a step runs it once, as you, with your permissions. There is an append-only staff audit log plus a separate log of everything Zrico did, with outcomes Did it, Asked a person and Didn't work. Stop everything is one switch, and shadow mode lets a person approve every message and change. On training, the underlying AI provider's policy is that Anthropic doesn't train its models on API data by default. Zrico doesn't have a separate written training policy of its own yet. Each company's data is kept separate with database row-level security, and you can set a hard monthly budget for AI spend.
How Zrico helps
You can check every answer above in the product during a demo at /demo. The full list of controls is at /ai/trust, the levels and limits are explained at /ai, and the security measures are listed at /security.